June 05, 2012, 10:33 AM — Last month's news that IBM would be pushing out Linux-on-Power based servers may very well have given you a sense of been-there-done-that, but the new focus on IBM's Power architecture is giving Big Blue another shot at taking on the Linux on x86 sector.
The question is, will the shot hit its big data target? Or are the new PowerLinux systems too expensive for their intended markets?
Coming off of its reported victory in the UNIX sector, where IBM's AIX purportedly has 52 percent of the market and growing, Big Blue is now shifting its attention to capturing a little bit more of the Linux/x86 market, according to Scott Handy, VP of IBM PowerLinux Strategy.
While he may not be a household name, Handy is a long-time presence within IBM's Linux hallways, and a familiar contact within Big Blue for various Linux-related initiatives over the years. So I felt pretty comfortable asking the really burning question in my mind: haven't we been down this Power-on-Linux road before?
Indeed we have, Handy explained. IBM's first foray into Linux systems on Power architecture were first launched in 2001, but the difference is that now IBM is much more focused on driving these systems into the Linux on x86 customer base, whereas before they were content to pickup customers migrating away from UNIX that still wanted to stick with the Power platform.
PowerLinux today has a shot at picking up the Linux/x86 clientele for three main reasons, Handy outlined. First, there is the attractiveness of IBM's pricing against that of VMware's, which Handy definitely sees as a target for Big Blue. Power's capabilities make it a better performer than VMware in the virtualization sector, Handy maintained.
The chip's features will also make it a better performer in a lot of services that traditionally fall to Linux on Intel systems, he added, such as Web, proxy, file, and mail servers. The PowerLinux platform will deliver more bang for the buck, IBM feels.
But the third area that really just fell into IBM's lap as they started preparing this redux PowerLinux initiative was big data, which started exploding even as they were in the midst of their 18-month planning. IBM's Research was already doing the legwork for Handy's team by determining how Hadoop could be optimized on Power--work that was going into the Watson supercomputer and the then-upcoming BigInsights platform.
This work was something that Handy was more than willing to capitalize on, and is a big part of why this new PowerLinux push is placing a big emphasis on big data.
So how hard would it be for a Linux/x86 customer to make the shift to a PowerLinux system? Well, remember the IBM Chiphopper project? (I sure do, and so does Handy.) This ISV-oriented porting program for Linux applications to get from x86 to Power systems helped port thousands of Linux applications to Power, Handy told me this week. And of those applications, only four needed to have lines of code changed. The rest just needed to be recompiled on Power and that's it.
And, more compellingly for big data customers in the Hadoop ecosystem, "since Java is an interpreted language, it doesn't even need to be compiled," Handy added.
IBM has also specifically adjusted the pricing for new PowerLinux systems to be comparable with that of Linux/x86 systems, which should help make the adjustment easier. Handy did not go into details in our discussion, but Timothy Prickett Morgan certainly has, with his usual excellent job of a pricing analysis on the new PowerLinux platform.
"To put it bluntly, IBM's pricing on the new PowerLinux 7R2 rack-based server, which is only certified to run Red Hat Enterprise Linux or SUSE Linux, is crazy stupid lower than what IBM i and AIX shops have to pay," Morgan wrote.
But looking at his pricing table, I have to wonder if the payoff for PowerLinux is worth the price tag, especially in big data. Both Hortonworks and Cloudera recommend mid-range systems for Hadoop deployments that run anywhere from $3,000-$7,000. The base price for a new PowerLinux 7R2 system is $19,560, according to Morgan's analysis.
You could argue that you're getting a lot more processing bang for the buck with one of these systems, but given that one of the big draws for Hadoop is the capability to linearly scale with inexpensive nodes, and that the storage on this particular model is only 600GB, I am having trouble reconciling the linear scalability of a super-powered node with mediocre storage.
There are surely costs included in Hadoop deployments from Cloudera, Hortonworks, or MapR that I haven't factored in over and above that base hardware cost, so take my comparison with a grain of salt. Still, IBM may be banking on its hardware rep more that it should and may miss the commodity-focused market that still revolves around Linux/x86.
Read more of Brian Proffitt's Zettatag and Open for Discussion blogs and follow the latest IT news at ITworld. Drop Brian a line or follow Brian on Twitter at @TheTechScribe. For the latest IT news, analysis and how-tos, follow ITworld on Twitter and Facebook.